Re: -fno-builtin not preventing __builtin___snprintf_chk in gcc 11.2.0-19ubuntu1

[Reid Wahl <nwahl@redhat.com>]

On Sun, Aug 21, 2022 at 5:53 PM Reid Wahl <nwahl@redhat.com> wrote:
>
> Hello,
>
> I'm hitting a strange issue with GCC while trying to wrap and mock
> snprintf() for unit testing purposes.
>
> I'm setting -fno-builtin and -fno-inline, but
> __builtin___snprintf_chk() is getting called instead of snprintf().
> The net effect is that my __wrap__snprintf() never gets called. Do you
> have any advice on how to prevent object size checking builtins from
> being used, without setting a lower optimization level? I expected
> -fno-builtin to take care of it.
>
> Note that out of my whole multi-distro test bed, I've only observed
> this issue on Ubuntu 22.04, Ubuntu 20.04, and Fedora 36 Power 9. I
> don't have direct access to the test machines, but I spun up an Ubuntu
> 22.04 reproducer. It uses gcc version 11.2.0-19ubuntu1.
>
> A known working Fedora 36 x86_64 system uses gcc version 12.1.1
> 20220507. On that system, snprintf() (or more accurately
> __wrap_snprintf()) gets called as expected, instead of the builtin.
> There are also RHEL 7 machines in the testbed, so it works fine on
> older versions as well.
>
> Minimal reproducer:
> ```
> # cat test.c
> #include <stdio.h>
> void func(const char *s)
> {
>     char buf[16];
>     snprintf(buf, 16, "%s", s);
>     printf("%s\n", buf);
> }
> int main(void)
> {
>     func("hello world");
> }
>
> # gcc -g -O2 -fno-builtin -fno-inline test.c -o test
> # gdb -q ./test
> Reading symbols from ./test...
> (gdb) b 5
> Breakpoint 1 at 0x10a0: file test.c, line 9.
> (gdb) r
> Starting program: /root/git/pacemaker/test
> [Thread debugging using libthread_db enabled]
> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>
> Breakpoint 1, main () at test.c:9
> 9    {
> (gdb) s
> 10        func("hello world");
> (gdb)
> func (s=s@entry=0x55555555600b "hello world") at test.c:3
> 3    {
> (gdb)
> 5        snprintf(buf, 16, "%s", s);
> (gdb)
> 0x00005555555551b9 in snprintf (__fmt=<optimized out>, __n=<optimized
> out>, __s=<optimized out>) at
> /usr/include/x86_64-linux-gnu/bits/stdio2.h:71
> 71      return __builtin___snprintf_chk (__s, __n, __USE_FORTIFY_LEVEL - 1,
> ```
>
> Thank you.
>
> --
> Regards,
>
> Reid Wahl (He/Him)
> Senior Software Engineer, Red Hat
> RHEL High Availability - Pacemaker

I figured out the Ubuntu issue. `man gcc` under `-O2`:

           NOTE: In Ubuntu 8.10 and later versions,
-D_FORTIFY_SOURCE=2 is set by default, and is activated when -O is set
to 2 or higher.  This enables additional compile-time and run-time
checks for
           several libc functions.  To disable, specify either
-U_FORTIFY_SOURCE or -D_FORTIFY_SOURCE=0.

I have no idea what's causing the unit test to fail on Fedora 36 Power
9 LE (the only system in the test bed that's still failing), but I
don't have a reproducer or enough information to ask a question. I'll
try to get one set up this week and go from there.

-- 
Regards,

Reid Wahl (He/Him)
Senior Software Engineer, Red Hat
RHEL High Availability - Pacemaker